fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56
  57 #ifdef CONFIG_KMOD
  58 #include <linux/kmod.h>
  59 #endif
  60
  61 int core_uses_pid;
  62 char core_pattern[128] = "core";
  63 int suid_dumpable = 0;
  64
  65 EXPORT_SYMBOL(suid_dumpable);
  66 /* The maximal length of core_pattern is also specified in sysctl.c */
  67
  68 static struct linux_binfmt *formats;
  69 static DEFINE_RWLOCK(binfmt_lock);
  70
  71 int register_binfmt(struct linux_binfmt * fmt)
  72 {
  73         struct linux_binfmt ** tmp = &formats;
  74
  75         if (!fmt)
  76                 return -EINVAL;
  77         if (fmt->next)
  78                 return -EBUSY;
  79         write_lock(&binfmt_lock);
  80         while (*tmp) {
  81                 if (fmt == *tmp) {
  82                         write_unlock(&binfmt_lock);
  83                         return -EBUSY;
  84                 }
  85                 tmp = &(*tmp)->next;
  86         }
  87         fmt->next = formats;
  88         formats = fmt;
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(register_binfmt);
  94
  95 int unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         struct linux_binfmt ** tmp = &formats;
  98
  99         write_lock(&binfmt_lock);
 100         while (*tmp) {
 101                 if (fmt == *tmp) {
 102                         *tmp = fmt->next;
 103                         write_unlock(&binfmt_lock);
 104                         return 0;
 105                 }
 106                 tmp = &(*tmp)->next;
 107         }
 108         write_unlock(&binfmt_lock);
 109         return -EINVAL;
 110 }
 111
 112 EXPORT_SYMBOL(unregister_binfmt);
 113
 114 static inline void put_binfmt(struct linux_binfmt * fmt)
 115 {
 116         module_put(fmt->module);
 117 }
 118
 119 /*
 120  * Note that a shared library must be both readable and executable due to
 121  * security reasons.
 122  *
 123  * Also note that we take the address to load from from the file itself.
 124  */
 125 asmlinkage long sys_uselib(const char __user * library)
 126 {
 127         struct file * file;
 128         struct nameidata nd;
 129         int error;
 130
 131         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 132         if (error)
 133                 goto out;
 134
 135         error = -EINVAL;
 136         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 137                 goto exit;
 138
 139         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 140         if (error)
 141                 goto exit;
 142
 143         file = nameidata_to_filp(&nd, O_RDONLY);
 144         error = PTR_ERR(file);
 145         if (IS_ERR(file))
 146                 goto out;
 147
 148         error = -ENOEXEC;
 149         if(file->f_op) {
 150                 struct linux_binfmt * fmt;
 151
 152                 read_lock(&binfmt_lock);
 153                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 154                         if (!fmt->load_shlib)
 155                                 continue;
 156                         if (!try_module_get(fmt->module))
 157                                 continue;
 158                         read_unlock(&binfmt_lock);
 159                         error = fmt->load_shlib(file);
 160                         read_lock(&binfmt_lock);
 161                         put_binfmt(fmt);
 162                         if (error != -ENOEXEC)
 163                                 break;
 164                 }
 165                 read_unlock(&binfmt_lock);
 166         }
 167         fput(file);
 168 out:
 169         return error;
 170 exit:
 171         release_open_intent(&nd);
 172         path_release(&nd);
 173         goto out;
 174 }
 175
 176 /*
 177  * count() counts the number of strings in array ARGV.
 178  */
 179 static int count(char __user * __user * argv, int max)
 180 {
 181         int i = 0;
 182
 183         if (argv != NULL) {
 184                 for (;;) {
 185                         char __user * p;
 186
 187                         if (get_user(p, argv))
 188                                 return -EFAULT;
 189                         if (!p)
 190                                 break;
 191                         argv++;
 192                         if(++i > max)
 193                                 return -E2BIG;
 194                         cond_resched();
 195                 }
 196         }
 197         return i;
 198 }
 199
 200 /*
 201  * 'copy_strings()' copies argument/environment strings from user
 202  * memory to free pages in kernel mem. These are in a format ready
 203  * to be put directly into the top of new user memory.
 204  */
 205 static int copy_strings(int argc, char __user * __user * argv,
 206                         struct linux_binprm *bprm)
 207 {
 208         struct page *kmapped_page = NULL;
 209         char *kaddr = NULL;
 210         int ret;
 211
 212         while (argc-- > 0) {
 213                 char __user *str;
 214                 int len;
 215                 unsigned long pos;
 216
 217                 if (get_user(str, argv+argc) ||
 218                                 !(len = strnlen_user(str, bprm->p))) {
 219                         ret = -EFAULT;
 220                         goto out;
 221                 }
 222
 223                 if (bprm->p < len)  {
 224                         ret = -E2BIG;
 225                         goto out;
 226                 }
 227
 228                 bprm->p -= len;
 229                 /* XXX: add architecture specific overflow check here. */
 230                 pos = bprm->p;
 231
 232                 while (len > 0) {
 233                         int i, new, err;
 234                         int offset, bytes_to_copy;
 235                         struct page *page;
 236
 237                         offset = pos % PAGE_SIZE;
 238                         i = pos/PAGE_SIZE;
 239                         page = bprm->page[i];
 240                         new = 0;
 241                         if (!page) {
 242                                 page = alloc_page(GFP_HIGHUSER);
 243                                 bprm->page[i] = page;
 244                                 if (!page) {
 245                                         ret = -ENOMEM;
 246                                         goto out;
 247                                 }
 248                                 new = 1;
 249                         }
 250
 251                         if (page != kmapped_page) {
 252                                 if (kmapped_page)
 253                                         kunmap(kmapped_page);
 254                                 kmapped_page = page;
 255                                 kaddr = kmap(kmapped_page);
 256                         }
 257                         if (new && offset)
 258                                 memset(kaddr, 0, offset);
 259                         bytes_to_copy = PAGE_SIZE - offset;
 260                         if (bytes_to_copy > len) {
 261                                 bytes_to_copy = len;
 262                                 if (new)
 263                                         memset(kaddr+offset+len, 0,
 264                                                 PAGE_SIZE-offset-len);
 265                         }
 266                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 267                         if (err) {
 268                                 ret = -EFAULT;
 269                                 goto out;
 270                         }
 271
 272                         pos += bytes_to_copy;
 273                         str += bytes_to_copy;
 274                         len -= bytes_to_copy;
 275                 }
 276         }
 277         ret = 0;
 278 out:
 279         if (kmapped_page)
 280                 kunmap(kmapped_page);
 281         return ret;
 282 }
 283
 284 /*
 285  * Like copy_strings, but get argv and its values from kernel memory.
 286  */
 287 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 288 {
 289         int r;
 290         mm_segment_t oldfs = get_fs();
 291         set_fs(KERNEL_DS);
 292         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 293         set_fs(oldfs);
 294         return r;
 295 }
 296
 297 EXPORT_SYMBOL(copy_strings_kernel);
 298
 299 #ifdef CONFIG_MMU
 300 /*
 301  * This routine is used to map in a page into an address space: needed by
 302  * execve() for the initial stack and environment pages.
 303  *
 304  * vma->vm_mm->mmap_sem is held for writing.
 305  */
 306 void install_arg_page(struct vm_area_struct *vma,
 307                         struct page *page, unsigned long address)
 308 {
 309         struct mm_struct *mm = vma->vm_mm;
 310         pte_t * pte;
 311         spinlock_t *ptl;
 312
 313         if (unlikely(anon_vma_prepare(vma)))
 314                 goto out;
 315
 316         flush_dcache_page(page);
 317         pte = get_locked_pte(mm, address, &ptl);
 318         if (!pte)
 319                 goto out;
 320         if (!pte_none(*pte)) {
 321                 pte_unmap_unlock(pte, ptl);
 322                 goto out;
 323         }
 324         inc_mm_counter(mm, anon_rss);
 325         lru_cache_add_active(page);
 326         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 327                                         page, vma->vm_page_prot))));
 328         page_add_new_anon_rmap(page, vma, address);
 329         pte_unmap_unlock(pte, ptl);
 330
 331         /* no need for flush_tlb */
 332         return;
 333 out:
 334         __free_page(page);
 335         force_sig(SIGKILL, current);
 336 }
 337
 338 #define EXTRA_STACK_VM_PAGES    20      /* random */
 339
 340 int setup_arg_pages(struct linux_binprm *bprm,
 341                     unsigned long stack_top,
 342                     int executable_stack)
 343 {
 344         unsigned long stack_base;
 345         struct vm_area_struct *mpnt;
 346         struct mm_struct *mm = current->mm;
 347         int i, ret;
 348         long arg_size;
 349
 350 #ifdef CONFIG_STACK_GROWSUP
 351         /* Move the argument and environment strings to the bottom of the
 352          * stack space.
 353          */
 354         int offset, j;
 355         char *to, *from;
 356
 357         /* Start by shifting all the pages down */
 358         i = 0;
 359         for (j = 0; j < MAX_ARG_PAGES; j++) {
 360                 struct page *page = bprm->page[j];
 361                 if (!page)
 362                         continue;
 363                 bprm->page[i++] = page;
 364         }
 365
 366         /* Now move them within their pages */
 367         offset = bprm->p % PAGE_SIZE;
 368         to = kmap(bprm->page[0]);
 369         for (j = 1; j < i; j++) {
 370                 memmove(to, to + offset, PAGE_SIZE - offset);
 371                 from = kmap(bprm->page[j]);
 372                 memcpy(to + PAGE_SIZE - offset, from, offset);
 373                 kunmap(bprm->page[j - 1]);
 374                 to = from;
 375         }
 376         memmove(to, to + offset, PAGE_SIZE - offset);
 377         kunmap(bprm->page[j - 1]);
 378
 379         /* Limit stack size to 1GB */
 380         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 381         if (stack_base > (1 << 30))
 382                 stack_base = 1 << 30;
 383         stack_base = PAGE_ALIGN(stack_top - stack_base);
 384
 385         /* Adjust bprm->p to point to the end of the strings. */
 386         bprm->p = stack_base + PAGE_SIZE * i - offset;
 387
 388         mm->arg_start = stack_base;
 389         arg_size = i << PAGE_SHIFT;
 390
 391         /* zero pages that were copied above */
 392         while (i < MAX_ARG_PAGES)
 393                 bprm->page[i++] = NULL;
 394 #else
 395         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 396         stack_base = PAGE_ALIGN(stack_base);
 397         bprm->p += stack_base;
 398         mm->arg_start = bprm->p;
 399         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 400 #endif
 401
 402         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 403
 404         if (bprm->loader)
 405                 bprm->loader += stack_base;
 406         bprm->exec += stack_base;
 407
 408         mpnt = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 409         if (!mpnt)
 410                 return -ENOMEM;
 411
 412         memset(mpnt, 0, sizeof(*mpnt));
 413
 414         down_write(&mm->mmap_sem);
 415         {
 416                 mpnt->vm_mm = mm;
 417 #ifdef CONFIG_STACK_GROWSUP
 418                 mpnt->vm_start = stack_base;
 419                 mpnt->vm_end = stack_base + arg_size;
 420 #else
 421                 mpnt->vm_end = stack_top;
 422                 mpnt->vm_start = mpnt->vm_end - arg_size;
 423 #endif
 424                 /* Adjust stack execute permissions; explicitly enable
 425                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 426                  * and leave alone (arch default) otherwise. */
 427                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 428                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 429                 else if (executable_stack == EXSTACK_DISABLE_X)
 430                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 431                 else
 432                         mpnt->vm_flags = VM_STACK_FLAGS;
 433                 mpnt->vm_flags |= mm->def_flags;
 434                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 435                 if ((ret = insert_vm_struct(mm, mpnt))) {
 436                         up_write(&mm->mmap_sem);
 437                         kmem_cache_free(vm_area_cachep, mpnt);
 438                         return ret;
 439                 }
 440                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 441         }
 442
 443         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 444                 struct page *page = bprm->page[i];
 445                 if (page) {
 446                         bprm->page[i] = NULL;
 447                         install_arg_page(mpnt, page, stack_base);
 448                 }
 449                 stack_base += PAGE_SIZE;
 450         }
 451         up_write(&mm->mmap_sem);
 452
 453         return 0;
 454 }
 455
 456 EXPORT_SYMBOL(setup_arg_pages);
 457
 458 #define free_arg_pages(bprm) do { } while (0)
 459
 460 #else
 461
 462 static inline void free_arg_pages(struct linux_binprm *bprm)
 463 {
 464         int i;
 465
 466         for (i = 0; i < MAX_ARG_PAGES; i++) {
 467                 if (bprm->page[i])
 468                         __free_page(bprm->page[i]);
 469                 bprm->page[i] = NULL;
 470         }
 471 }
 472
 473 #endif /* CONFIG_MMU */
 474
 475 struct file *open_exec(const char *name)
 476 {
 477         struct nameidata nd;
 478         int err;
 479         struct file *file;
 480
 481         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 482         file = ERR_PTR(err);
 483
 484         if (!err) {
 485                 struct inode *inode = nd.dentry->d_inode;
 486                 file = ERR_PTR(-EACCES);
 487                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 488                     S_ISREG(inode->i_mode)) {
 489                         int err = vfs_permission(&nd, MAY_EXEC);
 490                         file = ERR_PTR(err);
 491                         if (!err) {
 492                                 file = nameidata_to_filp(&nd, O_RDONLY);
 493                                 if (!IS_ERR(file)) {
 494                                         err = deny_write_access(file);
 495                                         if (err) {
 496                                                 fput(file);
 497                                                 file = ERR_PTR(err);
 498                                         }
 499                                 }
 500 out:
 501                                 return file;
 502                         }
 503                 }
 504                 release_open_intent(&nd);
 505                 path_release(&nd);
 506         }
 507         goto out;
 508 }
 509
 510 EXPORT_SYMBOL(open_exec);
 511
 512 int kernel_read(struct file *file, unsigned long offset,
 513         char *addr, unsigned long count)
 514 {
 515         mm_segment_t old_fs;
 516         loff_t pos = offset;
 517         int result;
 518
 519         old_fs = get_fs();
 520         set_fs(get_ds());
 521         /* The cast to a user pointer is valid due to the set_fs() */
 522         result = vfs_read(file, (void __user *)addr, count, &pos);
 523         set_fs(old_fs);
 524         return result;
 525 }
 526
 527 EXPORT_SYMBOL(kernel_read);
 528
 529 static int exec_mmap(struct mm_struct *mm)
 530 {
 531         struct task_struct *tsk;
 532         struct mm_struct * old_mm, *active_mm;
 533
 534         /* Notify parent that we're no longer interested in the old VM */
 535         tsk = current;
 536         old_mm = current->mm;
 537         mm_release(tsk, old_mm);
 538
 539         if (old_mm) {
 540                 /*
 541                  * Make sure that if there is a core dump in progress
 542                  * for the old mm, we get out and die instead of going
 543                  * through with the exec.  We must hold mmap_sem around
 544                  * checking core_waiters and changing tsk->mm.  The
 545                  * core-inducing thread will increment core_waiters for
 546                  * each thread whose ->mm == old_mm.
 547                  */
 548                 down_read(&old_mm->mmap_sem);
 549                 if (unlikely(old_mm->core_waiters)) {
 550                         up_read(&old_mm->mmap_sem);
 551                         return -EINTR;
 552                 }
 553         }
 554         task_lock(tsk);
 555         active_mm = tsk->active_mm;
 556         tsk->mm = mm;
 557         tsk->active_mm = mm;
 558         activate_mm(active_mm, mm);
 559         task_unlock(tsk);
 560         arch_pick_mmap_layout(mm);
 561         if (old_mm) {
 562                 up_read(&old_mm->mmap_sem);
 563                 BUG_ON(active_mm != old_mm);
 564                 mmput(old_mm);
 565                 return 0;
 566         }
 567         mmdrop(active_mm);
 568         return 0;
 569 }
 570
 571 /*
 572  * This function makes sure the current process has its own signal table,
 573  * so that flush_signal_handlers can later reset the handlers without
 574  * disturbing other processes.  (Other processes might share the signal
 575  * table via the CLONE_SIGHAND option to clone().)
 576  */
 577 static int de_thread(struct task_struct *tsk)
 578 {
 579         struct signal_struct *sig = tsk->signal;
 580         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 581         spinlock_t *lock = &oldsighand->siglock;
 582         struct task_struct *leader = NULL;
 583         int count;
 584
 585         /*
 586          * If we don't share sighandlers, then we aren't sharing anything
 587          * and we can just re-use it all.
 588          */
 589         if (atomic_read(&oldsighand->count) <= 1) {
 590                 BUG_ON(atomic_read(&sig->count) != 1);
 591                 exit_itimers(sig);
 592                 return 0;
 593         }
 594
 595         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 596         if (!newsighand)
 597                 return -ENOMEM;
 598
 599         if (thread_group_empty(tsk))
 600                 goto no_thread_group;
 601
 602         /*
 603          * Kill all other threads in the thread group.
 604          * We must hold tasklist_lock to call zap_other_threads.
 605          */
 606         read_lock(&tasklist_lock);
 607         spin_lock_irq(lock);
 608         if (sig->flags & SIGNAL_GROUP_EXIT) {
 609                 /*
 610                  * Another group action in progress, just
 611                  * return so that the signal is processed.
 612                  */
 613                 spin_unlock_irq(lock);
 614                 read_unlock(&tasklist_lock);
 615                 kmem_cache_free(sighand_cachep, newsighand);
 616                 return -EAGAIN;
 617         }
 618
 619         /*
 620          * child_reaper ignores SIGKILL, change it now.
 621          * Reparenting needs write_lock on tasklist_lock,
 622          * so it is safe to do it under read_lock.
 623          */
 624         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 625                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 626
 627         zap_other_threads(tsk);
 628         read_unlock(&tasklist_lock);
 629
 630         /*
 631          * Account for the thread group leader hanging around:
 632          */
 633         count = 1;
 634         if (!thread_group_leader(tsk)) {
 635                 count = 2;
 636                 /*
 637                  * The SIGALRM timer survives the exec, but needs to point
 638                  * at us as the new group leader now.  We have a race with
 639                  * a timer firing now getting the old leader, so we need to
 640                  * synchronize with any firing (by calling del_timer_sync)
 641                  * before we can safely let the old group leader die.
 642                  */
 643                 sig->tsk = tsk;
 644                 spin_unlock_irq(lock);
 645                 if (hrtimer_cancel(&sig->real_timer))
 646                         hrtimer_restart(&sig->real_timer);
 647                 spin_lock_irq(lock);
 648         }
 649         while (atomic_read(&sig->count) > count) {
 650                 sig->group_exit_task = tsk;
 651                 sig->notify_count = count;
 652                 __set_current_state(TASK_UNINTERRUPTIBLE);
 653                 spin_unlock_irq(lock);
 654                 schedule();
 655                 spin_lock_irq(lock);
 656         }
 657         sig->group_exit_task = NULL;
 658         sig->notify_count = 0;
 659         spin_unlock_irq(lock);
 660
 661         /*
 662          * At this point all other threads have exited, all we have to
 663          * do is to wait for the thread group leader to become inactive,
 664          * and to assume its PID:
 665          */
 666         if (!thread_group_leader(tsk)) {
 667                 /*
 668                  * Wait for the thread group leader to be a zombie.
 669                  * It should already be zombie at this point, most
 670                  * of the time.
 671                  */
 672                 leader = tsk->group_leader;
 673                 while (leader->exit_state != EXIT_ZOMBIE)
 674                         yield();
 675
 676                 /*
 677                  * The only record we have of the real-time age of a
 678                  * process, regardless of execs it's done, is start_time.
 679                  * All the past CPU time is accumulated in signal_struct
 680                  * from sister threads now dead.  But in this non-leader
 681                  * exec, nothing survives from the original leader thread,
 682                  * whose birth marks the true age of this process now.
 683                  * When we take on its identity by switching to its PID, we
 684                  * also take its birthdate (always earlier than our own).
 685                  */
 686                 tsk->start_time = leader->start_time;
 687
 688                 write_lock_irq(&tasklist_lock);
 689
 690                 BUG_ON(leader->tgid != tsk->tgid);
 691                 BUG_ON(tsk->pid == tsk->tgid);
 692                 /*
 693                  * An exec() starts a new thread group with the
 694                  * TGID of the previous thread group. Rehash the
 695                  * two threads with a switched PID, and release
 696                  * the former thread group leader:
 697                  */
 698
 699                 /* Become a process group leader with the old leader's pid.
 700                  * The old leader becomes a thread of the this thread group.
 701                  * Note: The old leader also uses this pid until release_task
 702                  *       is called.  Odd but simple and correct.
 703                  */
 704                 detach_pid(tsk, PIDTYPE_PID);
 705                 tsk->pid = leader->pid;
 706                 attach_pid(tsk, PIDTYPE_PID,  tsk->pid);
 707                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 708                 transfer_pid(leader, tsk, PIDTYPE_SID);
 709                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 710
 711                 tsk->group_leader = tsk;
 712                 leader->group_leader = tsk;
 713
 714                 tsk->exit_signal = SIGCHLD;
 715
 716                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 717                 leader->exit_state = EXIT_DEAD;
 718
 719                 write_unlock_irq(&tasklist_lock);
 720         }
 721
 722         /*
 723          * There may be one thread left which is just exiting,
 724          * but it's safe to stop telling the group to kill themselves.
 725          */
 726         sig->flags = 0;
 727
 728 no_thread_group:
 729         exit_itimers(sig);
 730         if (leader)
 731                 release_task(leader);
 732
 733         BUG_ON(atomic_read(&sig->count) != 1);
 734
 735         if (atomic_read(&oldsighand->count) == 1) {
 736                 /*
 737                  * Now that we nuked the rest of the thread group,
 738                  * it turns out we are not sharing sighand any more either.
 739                  * So we can just keep it.
 740                  */
 741                 kmem_cache_free(sighand_cachep, newsighand);
 742         } else {
 743                 /*
 744                  * Move our state over to newsighand and switch it in.
 745                  */
 746                 atomic_set(&newsighand->count, 1);
 747                 memcpy(newsighand->action, oldsighand->action,
 748                        sizeof(newsighand->action));
 749
 750                 write_lock_irq(&tasklist_lock);
 751                 spin_lock(&oldsighand->siglock);
 752                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 753
 754                 rcu_assign_pointer(tsk->sighand, newsighand);
 755                 recalc_sigpending();
 756
 757                 spin_unlock(&newsighand->siglock);
 758                 spin_unlock(&oldsighand->siglock);
 759                 write_unlock_irq(&tasklist_lock);
 760
 761                 if (atomic_dec_and_test(&oldsighand->count))
 762                         kmem_cache_free(sighand_cachep, oldsighand);
 763         }
 764
 765         BUG_ON(!thread_group_leader(tsk));
 766         return 0;
 767 }
 768
 769 /*
 770  * These functions flushes out all traces of the currently running executable
 771  * so that a new one can be started
 772  */
 773
 774 static void flush_old_files(struct files_struct * files)
 775 {
 776         long j = -1;
 777         struct fdtable *fdt;
 778
 779         spin_lock(&files->file_lock);
 780         for (;;) {
 781                 unsigned long set, i;
 782
 783                 j++;
 784                 i = j * __NFDBITS;
 785                 fdt = files_fdtable(files);
 786                 if (i >= fdt->max_fds)
 787                         break;
 788                 set = fdt->close_on_exec->fds_bits[j];
 789                 if (!set)
 790                         continue;
 791                 fdt->close_on_exec->fds_bits[j] = 0;
 792                 spin_unlock(&files->file_lock);
 793                 for ( ; set ; i++,set >>= 1) {
 794                         if (set & 1) {
 795                                 sys_close(i);
 796                         }
 797                 }
 798                 spin_lock(&files->file_lock);
 799
 800         }
 801         spin_unlock(&files->file_lock);
 802 }
 803
 804 void get_task_comm(char *buf, struct task_struct *tsk)
 805 {
 806         /* buf must be at least sizeof(tsk->comm) in size */
 807         task_lock(tsk);
 808         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 809         task_unlock(tsk);
 810 }
 811
 812 void set_task_comm(struct task_struct *tsk, char *buf)
 813 {
 814         task_lock(tsk);
 815         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 816         task_unlock(tsk);
 817 }
 818
 819 int flush_old_exec(struct linux_binprm * bprm)
 820 {
 821         char * name;
 822         int i, ch, retval;
 823         struct files_struct *files;
 824         char tcomm[sizeof(current->comm)];
 825
 826         /*
 827          * Make sure we have a private signal table and that
 828          * we are unassociated from the previous thread group.
 829          */
 830         retval = de_thread(current);
 831         if (retval)
 832                 goto out;
 833
 834         /*
 835          * Make sure we have private file handles. Ask the
 836          * fork helper to do the work for us and the exit
 837          * helper to do the cleanup of the old one.
 838          */
 839         files = current->files;         /* refcounted so safe to hold */
 840         retval = unshare_files();
 841         if (retval)
 842                 goto out;
 843         /*
 844          * Release all of the old mmap stuff
 845          */
 846         retval = exec_mmap(bprm->mm);
 847         if (retval)
 848                 goto mmap_failed;
 849
 850         bprm->mm = NULL;                /* We're using it now */
 851
 852         /* This is the point of no return */
 853         put_files_struct(files);
 854
 855         current->sas_ss_sp = current->sas_ss_size = 0;
 856
 857         if (current->euid == current->uid && current->egid == current->gid)
 858                 current->mm->dumpable = 1;
 859         else
 860                 current->mm->dumpable = suid_dumpable;
 861
 862         name = bprm->filename;
 863
 864         /* Copies the binary name from after last slash */
 865         for (i=0; (ch = *(name++)) != '\0';) {
 866                 if (ch == '/')
 867                         i = 0; /* overwrite what we wrote */
 868                 else
 869                         if (i < (sizeof(tcomm) - 1))
 870                                 tcomm[i++] = ch;
 871         }
 872         tcomm[i] = '\0';
 873         set_task_comm(current, tcomm);
 874
 875         current->flags &= ~PF_RANDOMIZE;
 876         flush_thread();
 877
 878         /* Set the new mm task size. We have to do that late because it may
 879          * depend on TIF_32BIT which is only updated in flush_thread() on
 880          * some architectures like powerpc
 881          */
 882         current->mm->task_size = TASK_SIZE;
 883
 884         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
 885                 suid_keys(current);
 886                 current->mm->dumpable = suid_dumpable;
 887                 current->pdeath_signal = 0;
 888         } else if (file_permission(bprm->file, MAY_READ) ||
 889                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 890                 suid_keys(current);
 891                 current->mm->dumpable = suid_dumpable;
 892         }
 893
 894         /* An exec changes our domain. We are no longer part of the thread
 895            group */
 896
 897         current->self_exec_id++;
 898
 899         flush_signal_handlers(current, 0);
 900         flush_old_files(current->files);
 901
 902         return 0;
 903
 904 mmap_failed:
 905         reset_files_struct(current, files);
 906 out:
 907         return retval;
 908 }
 909
 910 EXPORT_SYMBOL(flush_old_exec);
 911
 912 /*
 913  * Fill the binprm structure from the inode.
 914  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 915  */
 916 int prepare_binprm(struct linux_binprm *bprm)
 917 {
 918         int mode;
 919         struct inode * inode = bprm->file->f_path.dentry->d_inode;
 920         int retval;
 921
 922         mode = inode->i_mode;
 923         if (bprm->file->f_op == NULL)
 924                 return -EACCES;
 925
 926         bprm->e_uid = current->euid;
 927         bprm->e_gid = current->egid;
 928
 929         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
 930                 /* Set-uid? */
 931                 if (mode & S_ISUID) {
 932                         current->personality &= ~PER_CLEAR_ON_SETID;
 933                         bprm->e_uid = inode->i_uid;
 934                 }
 935
 936                 /* Set-gid? */
 937                 /*
 938                  * If setgid is set but no group execute bit then this
 939                  * is a candidate for mandatory locking, not a setgid
 940                  * executable.
 941                  */
 942                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 943                         current->personality &= ~PER_CLEAR_ON_SETID;
 944                         bprm->e_gid = inode->i_gid;
 945                 }
 946         }
 947
 948         /* fill in binprm security blob */
 949         retval = security_bprm_set(bprm);
 950         if (retval)
 951                 return retval;
 952
 953         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 954         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 955 }
 956
 957 EXPORT_SYMBOL(prepare_binprm);
 958
 959 static int unsafe_exec(struct task_struct *p)
 960 {
 961         int unsafe = 0;
 962         if (p->ptrace & PT_PTRACED) {
 963                 if (p->ptrace & PT_PTRACE_CAP)
 964                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 965                 else
 966                         unsafe |= LSM_UNSAFE_PTRACE;
 967         }
 968         if (atomic_read(&p->fs->count) > 1 ||
 969             atomic_read(&p->files->count) > 1 ||
 970             atomic_read(&p->sighand->count) > 1)
 971                 unsafe |= LSM_UNSAFE_SHARE;
 972
 973         return unsafe;
 974 }
 975
 976 void compute_creds(struct linux_binprm *bprm)
 977 {
 978         int unsafe;
 979
 980         if (bprm->e_uid != current->uid) {
 981                 suid_keys(current);
 982                 current->pdeath_signal = 0;
 983         }
 984         exec_keys(current);
 985
 986         task_lock(current);
 987         unsafe = unsafe_exec(current);
 988         security_bprm_apply_creds(bprm, unsafe);
 989         task_unlock(current);
 990         security_bprm_post_apply_creds(bprm);
 991 }
 992
 993 EXPORT_SYMBOL(compute_creds);
 994
 995 void remove_arg_zero(struct linux_binprm *bprm)
 996 {
 997         if (bprm->argc) {
 998                 unsigned long offset;
 999                 char * kaddr;
1000                 struct page *page;
1001
1002                 offset = bprm->p % PAGE_SIZE;
1003                 goto inside;
1004
1005                 while (bprm->p++, *(kaddr+offset++)) {
1006                         if (offset != PAGE_SIZE)
1007                                 continue;
1008                         offset = 0;
1009                         kunmap_atomic(kaddr, KM_USER0);
1010 inside:
1011                         page = bprm->page[bprm->p/PAGE_SIZE];
1012                         kaddr = kmap_atomic(page, KM_USER0);
1013                 }
1014                 kunmap_atomic(kaddr, KM_USER0);
1015                 bprm->argc--;
1016         }
1017 }
1018
1019 EXPORT_SYMBOL(remove_arg_zero);
1020
1021 /*
1022  * cycle the list of binary formats handler, until one recognizes the image
1023  */
1024 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1025 {
1026         int try,retval;
1027         struct linux_binfmt *fmt;
1028 #ifdef __alpha__
1029         /* handle /sbin/loader.. */
1030         {
1031             struct exec * eh = (struct exec *) bprm->buf;
1032
1033             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1034                 (eh->fh.f_flags & 0x3000) == 0x3000)
1035             {
1036                 struct file * file;
1037                 unsigned long loader;
1038
1039                 allow_write_access(bprm->file);
1040                 fput(bprm->file);
1041                 bprm->file = NULL;
1042
1043                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1044
1045                 file = open_exec("/sbin/loader");
1046                 retval = PTR_ERR(file);
1047                 if (IS_ERR(file))
1048                         return retval;
1049
1050                 /* Remember if the application is TASO.  */
1051                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1052
1053                 bprm->file = file;
1054                 bprm->loader = loader;
1055                 retval = prepare_binprm(bprm);
1056                 if (retval<0)
1057                         return retval;
1058                 /* should call search_binary_handler recursively here,
1059                    but it does not matter */
1060             }
1061         }
1062 #endif
1063         retval = security_bprm_check(bprm);
1064         if (retval)
1065                 return retval;
1066
1067         /* kernel module loader fixup */
1068         /* so we don't try to load run modprobe in kernel space. */
1069         set_fs(USER_DS);
1070
1071         retval = audit_bprm(bprm);
1072         if (retval)
1073                 return retval;
1074
1075         retval = -ENOENT;
1076         for (try=0; try<2; try++) {
1077                 read_lock(&binfmt_lock);
1078                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1079                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1080                         if (!fn)
1081                                 continue;
1082                         if (!try_module_get(fmt->module))
1083                                 continue;
1084                         read_unlock(&binfmt_lock);
1085                         retval = fn(bprm, regs);
1086                         if (retval >= 0) {
1087                                 put_binfmt(fmt);
1088                                 allow_write_access(bprm->file);
1089                                 if (bprm->file)
1090                                         fput(bprm->file);
1091                                 bprm->file = NULL;
1092                                 current->did_exec = 1;
1093                                 proc_exec_connector(current);
1094                                 return retval;
1095                         }
1096                         read_lock(&binfmt_lock);
1097                         put_binfmt(fmt);
1098                         if (retval != -ENOEXEC || bprm->mm == NULL)
1099                                 break;
1100                         if (!bprm->file) {
1101                                 read_unlock(&binfmt_lock);
1102                                 return retval;
1103                         }
1104                 }
1105                 read_unlock(&binfmt_lock);
1106                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1107                         break;
1108 #ifdef CONFIG_KMOD
1109                 }else{
1110 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1111                         if (printable(bprm->buf[0]) &&
1112                             printable(bprm->buf[1]) &&
1113                             printable(bprm->buf[2]) &&
1114                             printable(bprm->buf[3]))
1115                                 break; /* -ENOEXEC */
1116                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1117 #endif
1118                 }
1119         }
1120         return retval;
1121 }
1122
1123 EXPORT_SYMBOL(search_binary_handler);
1124
1125 /*
1126  * sys_execve() executes a new program.
1127  */
1128 int do_execve(char * filename,
1129         char __user *__user *argv,
1130         char __user *__user *envp,
1131         struct pt_regs * regs)
1132 {
1133         struct linux_binprm *bprm;
1134         struct file *file;
1135         int retval;
1136         int i;
1137
1138         retval = -ENOMEM;
1139         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1140         if (!bprm)
1141                 goto out_ret;
1142
1143         file = open_exec(filename);
1144         retval = PTR_ERR(file);
1145         if (IS_ERR(file))
1146                 goto out_kfree;
1147
1148         sched_exec();
1149
1150         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1151
1152         bprm->file = file;
1153         bprm->filename = filename;
1154         bprm->interp = filename;
1155         bprm->mm = mm_alloc();
1156         retval = -ENOMEM;
1157         if (!bprm->mm)
1158                 goto out_file;
1159
1160         retval = init_new_context(current, bprm->mm);
1161         if (retval < 0)
1162                 goto out_mm;
1163
1164         bprm->argc = count(argv, bprm->p / sizeof(void *));
1165         if ((retval = bprm->argc) < 0)
1166                 goto out_mm;
1167
1168         bprm->envc = count(envp, bprm->p / sizeof(void *));
1169         if ((retval = bprm->envc) < 0)
1170                 goto out_mm;
1171
1172         retval = security_bprm_alloc(bprm);
1173         if (retval)
1174                 goto out;
1175
1176         retval = prepare_binprm(bprm);
1177         if (retval < 0)
1178                 goto out;
1179
1180         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1181         if (retval < 0)
1182                 goto out;
1183
1184         bprm->exec = bprm->p;
1185         retval = copy_strings(bprm->envc, envp, bprm);
1186         if (retval < 0)
1187                 goto out;
1188
1189         retval = copy_strings(bprm->argc, argv, bprm);
1190         if (retval < 0)
1191                 goto out;
1192
1193         retval = search_binary_handler(bprm,regs);
1194         if (retval >= 0) {
1195                 free_arg_pages(bprm);
1196
1197                 /* execve success */
1198                 security_bprm_free(bprm);
1199                 acct_update_integrals(current);
1200                 kfree(bprm);
1201                 return retval;
1202         }
1203
1204 out:
1205         /* Something went wrong, return the inode and free the argument pages*/
1206         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1207                 struct page * page = bprm->page[i];
1208                 if (page)
1209                         __free_page(page);
1210         }
1211
1212         if (bprm->security)
1213                 security_bprm_free(bprm);
1214
1215 out_mm:
1216         if (bprm->mm)
1217                 mmdrop(bprm->mm);
1218
1219 out_file:
1220         if (bprm->file) {
1221                 allow_write_access(bprm->file);
1222                 fput(bprm->file);
1223         }
1224
1225 out_kfree:
1226         kfree(bprm);
1227
1228 out_ret:
1229         return retval;
1230 }
1231
1232 int set_binfmt(struct linux_binfmt *new)
1233 {
1234         struct linux_binfmt *old = current->binfmt;
1235
1236         if (new) {
1237                 if (!try_module_get(new->module))
1238                         return -1;
1239         }
1240         current->binfmt = new;
1241         if (old)
1242                 module_put(old->module);
1243         return 0;
1244 }
1245
1246 EXPORT_SYMBOL(set_binfmt);
1247
1248 #define CORENAME_MAX_SIZE 64
1249
1250 /* format_corename will inspect the pattern parameter, and output a
1251  * name into corename, which must have space for at least
1252  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1253  */
1254 static int format_corename(char *corename, const char *pattern, long signr)
1255 {
1256         const char *pat_ptr = pattern;
1257         char *out_ptr = corename;
1258         char *const out_end = corename + CORENAME_MAX_SIZE;
1259         int rc;
1260         int pid_in_pattern = 0;
1261         int ispipe = 0;
1262
1263         if (*pattern == '|')
1264                 ispipe = 1;
1265
1266         /* Repeat as long as we have more pattern to process and more output
1267            space */
1268         while (*pat_ptr) {
1269                 if (*pat_ptr != '%') {
1270                         if (out_ptr == out_end)
1271                                 goto out;
1272                         *out_ptr++ = *pat_ptr++;
1273                 } else {
1274                         switch (*++pat_ptr) {
1275                         case 0:
1276                                 goto out;
1277                         /* Double percent, output one percent */
1278                         case '%':
1279                                 if (out_ptr == out_end)
1280                                         goto out;
1281                                 *out_ptr++ = '%';
1282                                 break;
1283                         /* pid */
1284                         case 'p':
1285                                 pid_in_pattern = 1;
1286                                 rc = snprintf(out_ptr, out_end - out_ptr,
1287                                               "%d", current->tgid);
1288                                 if (rc > out_end - out_ptr)
1289                                         goto out;
1290                                 out_ptr += rc;
1291                                 break;
1292                         /* uid */
1293                         case 'u':
1294                                 rc = snprintf(out_ptr, out_end - out_ptr,
1295                                               "%d", current->uid);
1296                                 if (rc > out_end - out_ptr)
1297                                         goto out;
1298                                 out_ptr += rc;
1299                                 break;
1300                         /* gid */
1301                         case 'g':
1302                                 rc = snprintf(out_ptr, out_end - out_ptr,
1303                                               "%d", current->gid);
1304                                 if (rc > out_end - out_ptr)
1305                                         goto out;
1306                                 out_ptr += rc;
1307                                 break;
1308                         /* signal that caused the coredump */
1309                         case 's':
1310                                 rc = snprintf(out_ptr, out_end - out_ptr,
1311                                               "%ld", signr);
1312                                 if (rc > out_end - out_ptr)
1313                                         goto out;
1314                                 out_ptr += rc;
1315                                 break;
1316                         /* UNIX time of coredump */
1317                         case 't': {
1318                                 struct timeval tv;
1319                                 do_gettimeofday(&tv);
1320                                 rc = snprintf(out_ptr, out_end - out_ptr,
1321                                               "%lu", tv.tv_sec);
1322                                 if (rc > out_end - out_ptr)
1323                                         goto out;
1324                                 out_ptr += rc;
1325                                 break;
1326                         }
1327                         /* hostname */
1328                         case 'h':
1329                                 down_read(&uts_sem);
1330                                 rc = snprintf(out_ptr, out_end - out_ptr,
1331                                               "%s", utsname()->nodename);
1332                                 up_read(&uts_sem);
1333                                 if (rc > out_end - out_ptr)
1334                                         goto out;
1335                                 out_ptr += rc;
1336                                 break;
1337                         /* executable */
1338                         case 'e':
1339                                 rc = snprintf(out_ptr, out_end - out_ptr,
1340                                               "%s", current->comm);
1341                                 if (rc > out_end - out_ptr)
1342                                         goto out;
1343                                 out_ptr += rc;
1344                                 break;
1345                         default:
1346                                 break;
1347                         }
1348                         ++pat_ptr;
1349                 }
1350         }
1351         /* Backward compatibility with core_uses_pid:
1352          *
1353          * If core_pattern does not include a %p (as is the default)
1354          * and core_uses_pid is set, then .%pid will be appended to
1355          * the filename. Do not do this for piped commands. */
1356         if (!ispipe && !pid_in_pattern
1357             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1358                 rc = snprintf(out_ptr, out_end - out_ptr,
1359                               ".%d", current->tgid);
1360                 if (rc > out_end - out_ptr)
1361                         goto out;
1362                 out_ptr += rc;
1363         }
1364 out:
1365         *out_ptr = 0;
1366         return ispipe;
1367 }
1368
1369 static void zap_process(struct task_struct *start)
1370 {
1371         struct task_struct *t;
1372
1373         start->signal->flags = SIGNAL_GROUP_EXIT;
1374         start->signal->group_stop_count = 0;
1375
1376         t = start;
1377         do {
1378                 if (t != current && t->mm) {
1379                         t->mm->core_waiters++;
1380                         sigaddset(&t->pending.signal, SIGKILL);
1381                         signal_wake_up(t, 1);
1382                 }
1383         } while ((t = next_thread(t)) != start);
1384 }
1385
1386 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1387                                 int exit_code)
1388 {
1389         struct task_struct *g, *p;
1390         unsigned long flags;
1391         int err = -EAGAIN;
1392
1393         spin_lock_irq(&tsk->sighand->siglock);
1394         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1395                 tsk->signal->group_exit_code = exit_code;
1396                 zap_process(tsk);
1397                 err = 0;
1398         }
1399         spin_unlock_irq(&tsk->sighand->siglock);
1400         if (err)
1401                 return err;
1402
1403         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1404                 goto done;
1405
1406         rcu_read_lock();
1407         for_each_process(g) {
1408                 if (g == tsk->group_leader)
1409                         continue;
1410
1411                 p = g;
1412                 do {
1413                         if (p->mm) {
1414                                 if (p->mm == mm) {
1415                                         /*
1416                                          * p->sighand can't disappear, but
1417                                          * may be changed by de_thread()
1418                                          */
1419                                         lock_task_sighand(p, &flags);
1420                                         zap_process(p);
1421                                         unlock_task_sighand(p, &flags);
1422                                 }
1423                                 break;
1424                         }
1425                 } while ((p = next_thread(p)) != g);
1426         }
1427         rcu_read_unlock();
1428 done:
1429         return mm->core_waiters;
1430 }
1431
1432 static int coredump_wait(int exit_code)
1433 {
1434         struct task_struct *tsk = current;
1435         struct mm_struct *mm = tsk->mm;
1436         struct completion startup_done;
1437         struct completion *vfork_done;
1438         int core_waiters;
1439
1440         init_completion(&mm->core_done);
1441         init_completion(&startup_done);
1442         mm->core_startup_done = &startup_done;
1443
1444         core_waiters = zap_threads(tsk, mm, exit_code);
1445         up_write(&mm->mmap_sem);
1446
1447         if (unlikely(core_waiters < 0))
1448                 goto fail;
1449
1450         /*
1451          * Make sure nobody is waiting for us to release the VM,
1452          * otherwise we can deadlock when we wait on each other
1453          */
1454         vfork_done = tsk->vfork_done;
1455         if (vfork_done) {
1456                 tsk->vfork_done = NULL;
1457                 complete(vfork_done);
1458         }
1459
1460         if (core_waiters)
1461                 wait_for_completion(&startup_done);
1462 fail:
1463         BUG_ON(mm->core_waiters);
1464         return core_waiters;
1465 }
1466
1467 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1468 {
1469         char corename[CORENAME_MAX_SIZE + 1];
1470         struct mm_struct *mm = current->mm;
1471         struct linux_binfmt * binfmt;
1472         struct inode * inode;
1473         struct file * file;
1474         int retval = 0;
1475         int fsuid = current->fsuid;
1476         int flag = 0;
1477         int ispipe = 0;
1478
1479         binfmt = current->binfmt;
1480         if (!binfmt || !binfmt->core_dump)
1481                 goto fail;
1482         down_write(&mm->mmap_sem);
1483         if (!mm->dumpable) {
1484                 up_write(&mm->mmap_sem);
1485                 goto fail;
1486         }
1487
1488         /*
1489          *      We cannot trust fsuid as being the "true" uid of the
1490          *      process nor do we know its entire history. We only know it
1491          *      was tainted so we dump it as root in mode 2.
1492          */
1493         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1494                 flag = O_EXCL;          /* Stop rewrite attacks */
1495                 current->fsuid = 0;     /* Dump root private */
1496         }
1497         mm->dumpable = 0;
1498
1499         retval = coredump_wait(exit_code);
1500         if (retval < 0)
1501                 goto fail;
1502
1503         /*
1504          * Clear any false indication of pending signals that might
1505          * be seen by the filesystem code called to write the core file.
1506          */
1507         clear_thread_flag(TIF_SIGPENDING);
1508
1509         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1510                 goto fail_unlock;
1511
1512         /*
1513          * lock_kernel() because format_corename() is controlled by sysctl, which
1514          * uses lock_kernel()
1515          */
1516         lock_kernel();
1517         ispipe = format_corename(corename, core_pattern, signr);
1518         unlock_kernel();
1519         if (ispipe) {
1520                 /* SIGPIPE can happen, but it's just never processed */
1521                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1522                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1523                                corename);
1524                         goto fail_unlock;
1525                 }
1526         } else
1527                 file = filp_open(corename,
1528                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1529                                  0600);
1530         if (IS_ERR(file))
1531                 goto fail_unlock;
1532         inode = file->f_path.dentry->d_inode;
1533         if (inode->i_nlink > 1)
1534                 goto close_fail;        /* multiple links - don't dump */
1535         if (!ispipe && d_unhashed(file->f_path.dentry))
1536                 goto close_fail;
1537
1538         /* AK: actually i see no reason to not allow this for named pipes etc.,
1539            but keep the previous behaviour for now. */
1540         if (!ispipe && !S_ISREG(inode->i_mode))
1541                 goto close_fail;
1542         if (!file->f_op)
1543                 goto close_fail;
1544         if (!file->f_op->write)
1545                 goto close_fail;
1546         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1547                 goto close_fail;
1548
1549         retval = binfmt->core_dump(signr, regs, file);
1550
1551         if (retval)
1552                 current->signal->group_exit_code |= 0x80;
1553 close_fail:
1554         filp_close(file, NULL);
1555 fail_unlock:
1556         current->fsuid = fsuid;
1557         complete_all(&mm->core_done);
1558 fail:
1559         return retval;
1560 }